Homogeneous pixel selection algorithm for multitemporal InSAR

Abstract

Synthetic Aperture Radar Interferometry(InSAR)technique for distributed scatterers has become one of the current flavors.Inaccurate estimation of the covariance matrix is regarded as the most important source of error in such applications.The previous studies,named statistically homogeneous pixel selection algorithms,have demonstrated their values to refine the estimate accuracy for each target without loss of image resolution.In this paper,we review these methods globally under parametric and non-parametric statistical framework.The updated one will then be presented by integrating the advantages of the parametric statistics,and evaluated through synthetic and real data.Based on these algorithms,we finally introduce our SHPS-InSAR opensource toolbox,designed for homogeneous pixel selection and covariance matrix estimation.The toolbox includes almost all methods presented in this paper and tries to provide the optimal observable for the researchers in the field. 分布式雷达目标时序InSAR技术是目前InSAR形变监测领域的主流方向,其中同质样本选取是该技术的基础,其估计精度直接影响SAR影像分辨率与后续参数解算精度.本文在追踪最新研究进展之上,系统回顾了当今统计同质选点算法的优缺点.在参数与非参数两类统计方法的应用中,采用蒙特卡罗方法和真实数据验证定量比较算法差异以及适用场景.根据之前的研究结论,提出一种改进的最优参数统计同质样本选择方法.最后,论文介绍了团队研发的MATLAB开源工具包,涵盖了同质样本提取和时序InSAR协方差矩阵估计两部分内容,为InSAR科研人员和后续数据处理提供高质量、全分辨率的观测源.